Sleep problems with kernels >= 2.6.21 on powerpc

[Stefan Richter]

On  5 Sep, Andrew Morton wrote:
>> On Wed, 05 Sep 2007 19:47:42 +0200 Stefan Richter <stefanr at s5r6.in-berlin.de> wrote:
>> >>> Trying to free already-free IRQ 40
>> >>> pci_set_power_state(): 0002:20:0e.0: state=3, current state=5
>> >>> firewire_ohci: pci_set_power_state failed with -22<3>pci_device_suspend(): pci_suspend+0x0/0x9c [firewire_ohci]() returns -22
...
>> The old ohci1394.c used to ignore pci_set_power_state's return value.
>> In the pre 2.6.19-rc1 commit ea6104c22468239083857fa07425c312b1ecb424, I
>> added a fail-on-error.  This was toned down to a printk-on-err by pre
>> 2.6.19-rc4 commit 346f5c7ee7fa4ebee0e4c96415a7e59716bfa1d0.
> 
> OK.
> 
>> This was because of Benjamin Herrenschmidt's regression report:
>> http://lkml.org/lkml/2006/10/24/13
> 
> It's not clear _why_ pci_set_power_state() is failing.

The only -22 error path in pci_set_power_state is this:

	/* Validate current state:
	 * Can enter D0 from any state, but if we can only go deeper 
	 * to sleep if we're already in a low power state
	 */
	if (state != PCI_D0 && dev->current_state > state) {
		printk(KERN_ERR "%s(): %s: state=%d, current state=%d\n",
			__FUNCTION__, pci_name(dev), state, dev->current_state);
		return -EINVAL;
	} [...else...]

(There seems to be one "if" too many in the comment.)


dev->current_state was PCI_UNKNOWN, and this is not properly handled by
pci_set_power_state.  Some checks later, there is

	switch (dev->current_state) {
	case PCI_D0:
	case PCI_D1:
	case PCI_D2:
		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
		pmcsr |= state;
		break;
	case PCI_UNKNOWN: /* Boot-up */
		if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
		 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
			need_restore = 1;
		/* Fall-through: force to D0 */
	default:
		pmcsr = 0;
		break;
	}

But the PCI_UNKNOWN branch will never be reached because of the if ()
clause quoted above.

Also, this FireWire controller here surely had left the boot-up phase
already long ago when the reporter tried to suspend the machine.
-- 
Stefan Richter
-=====-=-=== =--= --=-=
http://arcgraph.de/sr/